Running & Debugging

Studio supports two execution modes, each suited to different tasks.

Execution modes

1. Chat — Conversational AI with dynamic tool use. No predefined steps; the agent decides what to do based on your messages.
2. Flow — Structured execution following the visual flow graph. Each node runs in sequence according to the edges and conditions you defined.

Running a flow

1. Click the Run button in the flow editor.
2. If the flow has input nodes, a dialog prompts you for values.
3. Nodes highlight as they execute and edges show data flow in real-time.
4. The chat panel displays agent responses and tool outputs alongside the visual execution.

Debugging tips

• Execution panel — Check the step-by-step progress to see which node is active and what data it produced.
• Tool call blocks — Expand them to inspect full input and output for each tool invocation.
• YAML source drawer — Verify the flow structure matches your intent.
• /status command — In chat sessions, use this to inspect current session state.
• CLI traces — Run astonish sessions show <id> --verbose from the terminal for detailed execution traces.

Error handling

When a node fails, the error is displayed inline on the canvas and in the execution panel. Two configuration options help manage failures:

• continue_on_error: true — Set on individual nodes to let the flow proceed past a failure.
• max_retries — Set on a node to automatically retry it a specified number of times before marking it as failed.